Energy-Constrained Inference Recovery Authority — Industrial Edge

The mandate

A new vision-model release increased accelerator demand at several factories, but deployment controls did not incorporate node-specific power caps, enclosure cooling, ambient conditions or competing automation workloads. Protective controls worked; nevertheless, inference dropped unpredictably and operators lost confidence in alerts during hot shifts. The edge-platform leader left after rollout was paused and accountability fragmented across central and site teams.

The interim must start within three weeks for nine months, directing fleet stabilisation, admission-control redesign, two seasonal load windows and successor induction. Permanent recruitment begins after a seventy-five-day site-envelope baseline. Five weeks are protected for overlap; the role will not extend to enterprise cloud migration, factory electrical expansion or unrelated model-product delivery.

Handover requires each production inference to resolve model and runtime version, accelerator, node, power and thermal envelope, ambient state, competing load, admission decision, degradation, operator display and business disposition. The successor must command unseen heatwave and site-islanding exercises, while plant and digital leaders accept documented hardware, telemetry and model-efficiency debt.

The authority may freeze releases, set workload admission and degradation policy, remove nodes from service, reprioritise edge capacity, direct the approved ₹75 crore recovery and appoint temporary platform leads. Plant Engineering retains electrical and cooling safety; Operations owns process action; model owners approve use cases. Permanent hiring, site capital and expenditure beyond delegation require executive approval.

Replacement of site power systems, redesign of production processes and development of new AI features are outside scope. The interim can require measurable model resource profiles and reliable hardware interfaces but cannot override protective equipment or operational-technology authority. Recovery is bounded to predictable computation inside declared energy, thermal and operating constraints across the installed edge fleet.

Why this seat is open

The release exposed a governance gap between centrally benchmarked models and physically constrained factory nodes, then the platform leader’s exit removed enterprise decision authority. Site teams can protect individual assets but cannot establish one admissible fleet policy. A temporary executive must restore confidence before the group resumes scaled deployment and hires for long-term optimisation.

What you will own

• Reconstruct failed or degraded inference across model, runtime, accelerator, node, electrical supply, cooling, ambient state and competing workload.
• Define workload resource envelopes and admission evidence across representative hardware generations, enclosures, shifts and site conditions.
• Decide which models and nodes may return, require constrained service, need profiling or must remain withdrawn.
• Command exercises for hot ambient, cooling degradation, power cap, accelerator throttling, telemetry loss and factory islanding.
• Establish graceful degradation so operators know which inference is current, delayed, reduced or unavailable before process action.
• Govern recovery against admitted-workload stability, energy per useful inference, alert continuity, site exceptions and recurrence trends.
• Transfer command through seasonal stress events and successor acceptance of fleet heterogeneity, telemetry and model-efficiency limitations.

Candidate qualifications

• Held executive edge-AI, industrial compute or accelerator-fleet authority across multiple operating plants with constrained power and cooling.
• Reconciled model and runtime demand against node-level electrical, thermal and competing operational workloads during live production.
• Stopped or degraded inference safely when benchmark performance failed to represent site-specific physical operating envelopes.
• Worked with Plant Engineering and operational-technology owners without treating software orchestration as authority over protective systems.
• Governed heterogeneous accelerators, telemetry quality and model efficiency through seasonal, enclosure and workload variation.
• Handed a recovered edge fleet to permanent leadership through heat, power-cap and connectivity exercises under production conditions.

Non-negotiables

• Available within three weeks for Pune leadership and frequent travel to factories during peak-temperature operating windows.
• Has governed physical energy and thermal limits of production inference fleets; cloud AI FinOps alone is insufficient.
• No undisclosed interest in accelerator, edge-server, cooling, vision-platform or automation suppliers inside the programme.
• Will preserve plant safety and operator clarity even where constrained admission reduces advertised AI coverage.

1. 49 words maximum. State your Pune availability and one inference release you constrained because its physical envelope was wrong.
2. 49 words maximum. How did you make graceful AI degradation unmistakable to operators during a thermal event?
3. 49 words maximum. Which unseen power-cap scenario would qualify the permanent leader before handover?